Title: An eye on buildings: Application of photogrammetric computer vision to the extraction building footprints in urban and suburban Perth
Mehdi Ravanbakhsh1
RMIT, School of Mathematical and Geospatial Sciences & CRC for Spatial Information
Paul Duncan, Rebecca Prior and David Elliott, Matthew Adam & Brendon McAtee
Location Products and Services, Landgate
, , , &
Executive Summary
Building footprints are key components of emergency preparedness and valuation of property. Local Government has an interest to ensure that buildings identified on a property have the appropriate building approvals to match. It has been a sought after dataset since the first capture was done in 2007 under a National Disaster Mitigation Program (NDMP) project.
The Building Footprints dataset, however, is maintained only on a best efforts level, and it is a manual capture process that is very time consuming and costly for Landgate. A cursory check by the Department of Fire and Emergency Services in the Perth Hills area found a significant number of omissions that minimise its utility in ensuring bushfire survivability of buildings in the Perth Hills.
Since 2007, significant developments in computer technology and the shift of Landgate from colour film to RGBi digital aerial photography capture make it worthwhile to look at automated extraction of building footprints. The ability to capture new building footprints and verify the accuracy of existing building footprints automatically is a much more cost effective way to provide this dataset to Landgate’s customers.
Previous research has led to the development of feature extraction software that uses a combination of imagery products to extract data, but has a high reliance on LiDAR. LiDAR is expensive to capture and is not a component of Landgate’s standard capture program, therefore past research outcomes have limited value for Landgate.
In this research, an approach for accurate and rapid extraction of building outlines in 2D space was developed that uses overlapping aerial imagery only. The proposed approach was tested in 1x1 sq km areas in urban and suburban Perth and quality measure were estimated against reference data plotted manually.